On this page
AI grounding is an AI system using your pages as input to an answer it is writing, and from February 1, 2027, Bing will honor new AI Training and AI Grounding preferences that replace NoArchive and NoCache (Bing). Sites that use those old tags need to set the new preferences by March 1, 2027 to keep their restrictions, and Google plans to read the shared AI training opt-out as a Google-Extended block once it supports the signal, which it aims to do in 2027; that would also end grounding in Gemini Apps unless a site allows Google-Extended by name (Google). A health site that wants to be cited should keep grounding allowed on its public pages. Curve Compliance then shows which AI assistants send visitors, recorded on its own servers under a BAA.
Book a call. Want to see which AI assistants send patients to your site? Curve Compliance records the referring site on its own servers, under a BAA, so assistants such as chatgpt.com and perplexity.ai show up as sources in your reports. Curve's team does the setup, typically live in about a week. Book a call with Curve.
What Microsoft and Google announced on October 30, 2026
Microsoft says it “is building on the shared AI Training preference and adding a separate AI Grounding preference.” The preferences cover content “used for training or to inform generated responses, including web-grounded responses in Bing, Copilot, and Microsoft 365,” and “These preferences will replace existing Microsoft preferences NoArchive and NoCache for training and grounding” (Bing).
The same day, Google announced “plans to support a shared AI training control in robots.txt” and said: “We aim to support this new control along with other AI providers in 2027” (Google). Its crawler documentation adds that “Google intends to support the AIPREF shared machine-readable training preference signal” “in early 2027” (Google).
Training vs grounding, in Microsoft's words
“AI Training covers using an asset to modify the learned parameters of a generative AI model.” “AI Grounding covers using content as input to help inform AI-generated responses. Indexing and ranking pages in traditional search results is not grounding.” Microsoft adds: “Training and grounding are independent preferences” and “Both default to Allow when no applicable preference is published” (Bing).
| Preference | Where you set it | Choices | Default |
|---|---|---|---|
| AI Training | robots.txt for a site or path, using User-Agent: __usage_ai-training, or a page-level meta tag or HTTP header | Allow or Disallow | Allow |
| AI Grounding | Page level only, as a meta tag or HTTP header such as X-Robots-Tag: Bingbot: ai-grounding:disallow | Allow, DisallowExtended or Disallow | Allow |
Microsoft says “There is no site-wide robots.txt grounding preference,” and that “These usage preferences do not change eligibility for traditional Bing search results, subject to normal crawl and indexing requirements” (Bing).
Bing's dates
- December 15, 2026. “temporary grounding transition preferences open in Bing Webmaster Tools”; Microsoft says “this transition applies only to and is only available for sites using NoArchive or NoCache as of that date” (Bing).
- February 1, 2027. “Starting February 1, 2027, Microsoft will honor the new preferences as it crawls” (Bing).
- March 1, 2027. Complete the steps by then, because “after that date, NoArchive and NoCache will no longer be read as a preference for training or grounding” (Bing).
- September 30, 2027. “temporary preferences expire” (Bing).
Two details catch sites out. “Publishing a setting is not the same as activating it. Microsoft must crawl the page to discover the setting.” And for sites still on the old tags: “If you do not record a temporary grounding preference by March 1 in Bing Webmaster Tools, pages using NoArchive or NoCache default to Allow for grounding until Microsoft recrawls them and discovers your new AI Grounding preference” (Bing).
Disallow vs DisallowExtended for grounding
Disallow keeps a page out of grounding. DisallowExtended “limits grounding use to Microsoft-operated experiences such as AI Mode, AI Overview, Copilot, and Microsoft 365, and excludes grounding services Microsoft provides to third parties, such as Web IQ and grounding through API access in Foundry.” The tag form is ai-grounding:disallowextended. Microsoft is direct about the trade: “If you select DisallowExtended or Disallow for grounding, you will see reduced brand visibility in AI experiences,” and it adds that allowing grounding or training does not promise inclusion, traffic or payment (Bing).
For a clinic, grounding is how an assistant can name your practice and link to it when someone asks where to get care. Blocking it on public pages trades that away. For how Copilot and Bing answers pick sources, see the Bing and Copilot visibility playbook.
Google's shared training signal and the Google-Extended catch
Google says: “When a website uses this policy token to opt out of training in its robots.txt file (and Google-Extended is not explicitly specified), Google plans to interpret that as if the publisher had disallowed Google-Extended.” It adds: “Websites that specifically allow Google-Extended are eligible to appear in responses and receive clicks via grounding in Gemini Apps” (Google).
That matters because Google-Extended controls both Gemini training and “grounding (providing content from the Google Search index to the model at prompt time to improve factuality and relevancy)” in Gemini Apps and Grounding with Google Search on Vertex AI. Google also says “Google-Extended does not impact a site's inclusion in Google Search nor is it used as a ranking signal in Google Search” (Google). So with Google, training and grounding travel together under one token. Once Google supports the shared signal, a site that adds the shared training opt-out for every provider, and says nothing about Google-Extended, would also drop out of Gemini Apps grounding. If you want Gemini to keep citing you, name Google-Extended in robots.txt and allow it.
A setup for a clinic site
| Page type | AI Training | AI Grounding (Bing) and Google-Extended | Why |
|---|---|---|---|
| Service, condition and education pages | Your choice | Allow | These are the pages assistants cite when people ask where to get care. |
| Location and provider pages | Your choice | Allow | They answer near-me and provider questions. |
| Blog and guides | Your choice | Allow | Citations here build trust in the practice. |
| Booking, intake, portal and logged-in pages | Not the right control | Not the right control | Keep them out of public reach with authentication; see below. |
Microsoft is clear about the limit: “These settings govern use of content, not crawl access. An Allow setting does not override a robots.txt rule blocking Bingbot from crawling” (Bing). A usage preference is a request about how content is used once it is fetched. Patient portals and booking flows need real access controls, and robots.txt is a request, not a lock; see AI crawlers on healthcare websites: what to allow. If a public page could show patient details, that is a HIPAA question rather than a crawler setting, so talk to your counsel. For the rest of a site review, see the HIPAA compliant website checklist, and for this month's other changes, health marketing compliance news for November 2026.
How Curve Compliance helps
Once your public pages are open to grounding, the next question is whether AI answers send people who book. Curve Compliance records the referring hostname on its own servers, under a BAA, so chatgpt.com and perplexity.ai show up as sources in your reports, and Curve Analyst answers questions about that traffic in plain language from your own data. See tracking ChatGPT and Perplexity referrals without GA4.
For paid traffic, Curve replaces browser ad pixels with one script and sends conversions server-side to Meta, Google Ads, TikTok, Microsoft Advertising and LinkedIn, with a fixed list of fields for each, and strips PHI before data reaches an ad platform. Curve's free website scan lists the pixels and trackers on a page. Curve is SOC 2 Type II, signs a BAA on every plan, and its team does the setup, typically live in about a week.
Sources, checked November 2, 2026: Bing, Your content, your choice: clearer preferences for training and grounding; Google Search Central, Supporting a shared AI training control for websites; Google, common crawlers and Google-Extended.
Frequently Asked Questions
What is AI grounding?
Microsoft defines AI grounding as “using content as input to help inform AI-generated responses” and says indexing and ranking pages in traditional search results is not grounding (Bing).
Will blocking AI training hurt my Bing rankings?
Microsoft says the training and grounding preferences “do not change eligibility for traditional Bing search results,” subject to normal crawl and indexing requirements (Bing).
What happens to NoArchive and NoCache after March 1, 2027?
Bing stops reading them as a preference for training or grounding. Sites that relied on them should publish AI Training and AI Grounding preferences, and record a temporary grounding preference in Bing Webmaster Tools, before then (Bing).
Does DisallowExtended keep my site in Copilot answers?
It limits grounding to Microsoft-operated experiences such as Copilot, AI Mode and AI Overview, and excludes third-party grounding services, but Microsoft still warns of reduced brand visibility in AI experiences (Bing).
If I opt out of AI training with the shared token, can Gemini still cite my site?
Once Google supports the shared signal, which it aims to do in 2027, only if you also allow Google-Extended by name. Without that, Google plans to treat the opt-out as a Google-Extended block, which covers grounding in Gemini Apps (Google; Google).
Should a healthcare site allow AI grounding?
On public service, location and education pages, yes, if you want to be cited: Microsoft says blocking grounding reduces brand visibility in AI experiences (Bing). Patient-facing logged-in pages need access controls instead.
Do these settings keep AI agents out of patient portals?
No. Microsoft says they “govern use of content, not crawl access” (Bing). Portals and booking flows need authentication; see what to allow.
Talk to Curve Compliance
Book a call. Opening your public pages to AI answers? Curve Compliance shows which assistants send visitors, measures bookings server-side, strips PHI before data reaches an ad platform and signs a BAA on every plan. Curve's team does the setup, typically live in about a week. Book a call with Curve.